8.6 Attention as precision

What about attention? The classical view treats attention as a selection mechanism — the brain chooses which inputs to process more fully. In the predictive-coding framework, attention has a precise mathematical interpretation: it is the modulation of precision on prediction errors. Attending to a stimulus increases the precision assigned to its prediction errors, making them more influential in updating the brain’s representation. Inattention decreases precision and lets prior expectations dominate.

This reframes the famous cocktail-party effect — the ability to follow one conversation while ignoring others at a noisy party. Attention to one talker increases the precision of that talker’s prediction errors; competing talkers’ prediction errors get downweighted; the listener’s percept follows the attended talker. Neurally, this is implemented by gain modulation in auditory cortex, observable in fMRI and ECOG: attending to one of two simultaneous voices increases the cortical response to that voice and decreases the response to the unattended one. The same input drives different cortical responses, depending on what the listener is paying attention to.

Two listeners at the same party can hear different conversations, from the same acoustic mixture, because their attention — and therefore their precision allocation — differs.